Subverses is a simple CLI application to translate YouTube video subtitles to a different language. It uses the YouTube API to download the subtitles and the OpenAI API to translate them. If no subtitles are available for the video, the application will try to transcribe the audio and then translate it.
You can install Subverses directly from GitHub:
pip install git+https://github.com/your-username/subverses.gitOr install it in development mode:
git clone https://github.com/your-username/subverses.git
cd subverses
pip install -e .After installation, you can use the subverses command directly:
subverses youtube https://www.youtube.com/watch?v=...Or translate an existing subtitle file:
subverses srt path/to/subtitle.srtThe YouTube command will download and create files in a dedicated subdirectory under data directory.
For your convenience, the subdirectory will be named after the video title.
./subverses --help
Usage: subverses [OPTIONS] YOUTUBE_URL
Reasoning questions generation tool
Arguments:
YOUTUBE_URL URL of the YouTube video to download. [required]
Options:
--whisper-prompt TEXT Prompt for the whisper model. See https://co
okbook.openai.com/examples/whisper_prompting
_guide for more information.
--translate-additional-prompt TEXT
Additional prompt for the translation model.
--whisper-model TEXT Transcription model name. [default:
whisper-1]
--gpt-model TEXT Translation model name. [default:
gpt-3.5-turbo]
--dont-transcribe-audio / --no-dont-transcribe-audio
Fail if there is no manual transcript
available. [default: dont-transcribe-audio]
--force-transcription-from-audio / --no-force-transcription-from-audio
Force transcription from audio file even if
downloading manual transcript is possible.
[default: no-force-transcription-from-audio]
--start-transcription-segment INTEGER
Start transcription from this segment
number. [default: 0]
--translate-from TEXT Translate from language. Use two letter ISO
639-1 country code. [default: en]
--translate-to TEXT Translate to language. Use full language
name. [default: Polish]
--data-dir PATH Directory to store the downloaded data.
[default: data]
--download-max-retries INTEGER Maximum number of retries for downloading.
[default: 2]
--skip-existing / --no-skip-existing
When downloading audio and video, skip if
the file already exists. [default: skip-
existing]
--min-silence-len-sec INTEGER The minimum length of silence to detect,
when audio splitting is needed (in seconds).
[default: 2]
--silence-threshold INTEGER The silence threshold used for audio
splitting. It should be negative integer in
range -60 to -5 dB. [default: -30]
--verbose / --no-verbose Verbose output. [default: no-verbose]
--help Show this message and exit.
The following options can be set int the .env configuration file (please use CAPITAL_LETTERS for the option names in the .env file):
# CLI options
youtube_url: str
whisper_prompt: str | None
translate_additional_prompt: str | None
whisper_model: str
gpt_model: str
dont_transcribe_audio: bool
force_transcription_from_audio: bool
start_transcription_segment: int
translate_from: str = "en"
translate_to: str = "Polish"
data_dir: Path
download_max_retries: int
skip_existing: bool
min_silence_len_sec: int
silence_threshold: int
verbose: bool
# .env only options
openai_api_key: str | None = None
openai_organization: str | None = None
openai_base_url: str | None = None
whisper_openai_timeout: float | None | NotGiven = NOT_GIVEN
whisper_openai_max_retries: int | None = 2
For example to set the OPENAI_API_KEY, create a .env file in the root directory of the project and add the following line:
OPENAI_API_KEY=sk-...
To set the translate_to option to "German" add the following line to the .env file:
TRANSLATE_TO=German
Check the output of the subverses --help command for the full list of options.